Grid Multi-Body: Fix Row Scroll Pinning for Thumb Dragging

Input from Gemini 3.1 Pro (Antigravity):

✦ Hey Tobi, have a good night! I've spent the session analyzing the jitter and I agree that updating the Playwright e2e tests (GridRowPinning.spec.mjs, GridThumbDrag.spec.mjs) for the new Multi-Body architecture is our absolute best next step. It will give us frame-by-frame DOM traces to see exactly what's fighting us.

Here is a summary of my findings to review tomorrow:

1. The Mac OS Overlay Scrollbar Trap: In the tests you just ran, isThumbDragging often stayed false because the strict boundary detection failed on Mac OS overlay scrollbars (which have clientWidth === offsetWidth and don't take up physical layout space, so offsetX > clientWidth is false).
2. The Loss of scrollbarNode: Before the Epic, GridRowScrollPinning listened for mousedown on the explicit Neo.component.ScrollBar DOM node. Now, because it's a native overflow: auto scrollbar, it listens on the wrapperNode itself. This makes reliably detecting "is the user actually dragging the thumb instead of just clicking inside the grid?" incredibly difficult and prone to false positives/negatives.
3. The GridDragScroll Conflict: You were spot on. GridDragScroll registers to the bodyId + __wrapper (which is the EXACT same wrapper node that pinning listens to). It intercepts drags and modifies scrollTop artificially. This physics loop is almost certainly colliding with the pinning offset logic.
4. The Synchronous Math: I traced the VDOM update cycle in DeltaUpdates.mjs. GridRowScrollPinning.onDeltaUpdate correctly catches the event before the Main Thread applies the DOM mutations. This means the Math is perfectly synchronous and safe from 1-frame snap-backs (which is why the old version worked flawlessly). The architecture is sound; the regression is 100% in the DOM event detection layer (mousedown/isThumbDragging) and layout overlap.

Next Steps:

1. Update the Playwright e2e tests for the new Multi-Body wrapper DOM structure.
2. Use the Playwright Trace Viewer to see the exact frame-by-frame state of --grid-row-pin-offset vs scrollTop during the jitter.
3. Implement a robust, cross-platform way to detect native scrollbar drags, potentially by isolating the scrollbar area or refining the onMouseDown coordinates.

Input from Gemini 3.1 Pro (Antigravity):

✦ As requested, I have analyzed the previous debug state and pushed a clean-slate commit directly to dev.

I also re-implemented the strict edge detection logic for GridRowScrollPinning.mjs that was lost when we reset the addon yesterday. The new bounded check logic now successfully differentiates between standard hardware scrollbars and Mac OS overlay scrollbars. GridDragScroll was also updated with the identical bounded logic so that the two addons will no longer fight when the user grabs the native scrollbar track.

We have a perfectly clean baseline for when we start investigating the Playwright traces!

Input from Gemini 3.1 Pro (Antigravity):

✦ Playwright E2E tests have been stabilized against the new Multi-Body wrapper architecture. The strict mode violations have been resolved by excluding .neo-container from the .neo-grid-body-wrapper selector, and the boundary detection for mousedown has been accurately mapped to rect.right - 5. This provides a baseline telemetry capture. Moving forward to a new ticket to track and synthesize the 50px/sec manual drag jitter.